← ThisIsTheWay.to

Build with AI Agents

How Agents Fail

Common patterns of agent failures and how to address them

Most agent failures fall into a small number of patterns. Recognizing the pattern helps you choose the right primitive to address it.

Click any card to see typical fixes for that pattern.

Truth-misses (Hallucinations)

Agent confidently invents things that don't exist

Symptoms

  • Invents API fields, file paths, or CLI flags
  • Sounds plausible but doesn't match reality
  • Confident tone despite being wrong

Reactivity-misses

Code compiles but doesn't work at runtime

Symptoms

  • Passes type checking but fails at runtime
  • Misses implicit rules about system behavior
  • Works in isolation but breaks in context

Integration-misses

Doesn't understand what runs where

Symptoms

  • Confuses server vs client code
  • Mixes up CI vs local behavior
  • Doesn't understand mock vs real boundaries

Taste-misses

Technically correct but aesthetically wrong

Symptoms

  • DOM assertions pass but UI looks off
  • Spacing, hierarchy, or affordances are wrong
  • Functional but not polished

Process-misses

Claims done without actually finishing

Symptoms

  • Skips running tests
  • Doesn't update related docs
  • Claims completion prematurely

The Loop

  1. 1

    Miss

    Capture what went wrong

  2. 2

    Diagnosis

    What didn't it see?

  3. 3

    Primitive

    Which lever to pull

  4. 4

    Artifact

    Encode the fix

  5. 5

    Gate

    Enforce when ready